Reported beings,
across the record

Episodes: the existing name/alias scanner searches the loaded machine-caption corpus. Whole-word matches use its shared lexicon; generic words are excluded and overlapping names can be assigned to another record. Mentioned categories can be disputed, fictional, quoted, or discussed without a sighting. Missed aliases and caption errors can undercount.

Alternate uploads and cuts mapped to the same canonical episode count once. The header counts the union of matching episodes across categories, not the sum of the cards; its denominator is the loaded canonical corpus, not every public episode.

Cited passage links: distinct loaded transcript blocks with a resolvable timestamp cited by that category, including nested citations. Duplicate links to the same upload and block count once. Different cuts can retain separate passage links; their timestamps are not interchangeable. A working link does not verify its quotation or claim. External sources and missing/out-of-range timestamps are not counted.

Linked events: existing event records linked from a category, its chronology, or an event’s backlink. Duplicates and missing records are excluded. An event record is not necessarily an encounter. Coverage is not an aggregate truth score.